diff options
author | Reka Norman <rekanorman@google.com> | 2022-09-16 10:39:49 +1000 |
---|---|---|
committer | Felix Held <felix-coreboot@felixheld.de> | 2022-09-24 01:23:09 +0000 |
commit | ee63b44c47cbb8c81ce0e173bbc12129c3edbe71 (patch) | |
tree | e96ed996387389db2534967646a50c72b45eb95a /src/mainboard/lenovo/t530/gma-mainboard.ads | |
parent | 1df1cf994aa9ebdd9d51103d2d618f6aa66df2d6 (diff) |
drivers/mrc_cache: Compare hashes instead of full data
The current MRC cache update process is slow (28 ms on nissa), because
cbmem is not cached in romstage. Specifically, the new MRC data returned
by the FSP is stored in the FSP reserved memory in cbmem, so operations
on the new data (computing the checksum, comparing to the old data) are
slow.
Replace the data checksum in the MRC header with a hash, and compare
hashes instead of comparing the full data. This has two benefits:
1. The xxhash function is faster than computing an IP checksum (4 ms vs
14 ms on uncached data on nissa).
2. There's no need to memcmp() the full MRC data, which takes 14 ms on
nissa.
Before:
550:starting to load ChromeOS VPD 867,930 (4,664)
3:after RAM initialization 896,020 (28,090)
4:end of romstage 906,274 (10,254)
After:
550:starting to load ChromeOS VPD 864,820 (4,649)
3:after RAM initialization 869,652 (4,831)
4:end of romstage 879,909 (10,257)
BUG=b:242667207
TEST=Check that MRC caching still works as expected on nissa. Corrupt
the MRC cache and check that memory is retrained.
Change-Id: I1b7848d1d05e555b61e0f1cb605550dfe3449c6d
Signed-off-by: Reka Norman <rekanorman@chromium.org>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/67670
Reviewed-by: Julius Werner <jwerner@chromium.org>
Reviewed-by: Tim Wawrzynczak <twawrzynczak@chromium.org>
Reviewed-by: Eric Lai <eric_lai@quanta.corp-partner.google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Diffstat (limited to 'src/mainboard/lenovo/t530/gma-mainboard.ads')
0 files changed, 0 insertions, 0 deletions